SAN FRANCISCO — Microsoft brought together its chief executive Satya Nadella and Nvidia boss Jensen Huang on a stage here Wednesday. The two men shook hands over new hardware. They talked about agents that work while their owners sleep.

The event marked the company’s first major Windows presentation in two years. It delivered concrete products and a clearer vision than many expected. No vaporware promises. Real prices. Real ships dates. And a pointed challenge to Apple.

The Surface Laptop Ultra starts at $2,599. It carries Nvidia’s RTX Spark chip. Configurations climb past $5,000. One model with 128 gigabytes of unified memory and a 20-core processor lists for $5,899. Pre-orders opened immediately. Devices reach customers October 16. Ars Technica reported.

That price draws sharp breath. Memory costs have climbed. A chip shortage pushed component prices higher across the industry. Yet the machines target developers and creators who currently pay cloud bills for every inference. Local execution changes the math. No token charges for routine work.

The RTX Spark combines a Grace CPU with a Blackwell RTX GPU. It delivers up to one petaflop of AI performance. Unified memory reaches 128 gigabytes. Models exceeding 120 billion parameters run locally. Microsoft claims the laptop returns AI prompts 2.1 times faster than a 16-inch MacBook Pro with Apple’s M5 Pro chip. Image generation runs 4.3 times faster in company tests. Video generation hits six times the speed of the comparable Apple system. Bloomberg noted the benchmarks.

A companion device pushes further. The Surface RTX Spark Dev Box costs $5,999. It ships in November. The anodized aluminum chassis doubles as heatsink. It arrives pre-loaded with developer tools — Visual Studio Code, GitHub Copilot, WSL. One petaflop lives inside the compact box. Nvidia previewed its DGX Station for Windows alongside it. That system aims at deskside supercomputing with far higher memory and performance.

But hardware alone never tells the full story. Microsoft spent equal time on software. It calls the approach Hybrid Intelligence. Copilot gains permission-based access to local files. It reads documents, renames them, zips folders, drafts emails. All without leaving the machine when possible.

A redesigned Copilot app splits into three views. Home offers a place to chat with models. Code lets users prompt their way to widgets and full Windows applications. Autopilot oversees agents. Users watch tasks, approve actions, intervene when needed. The interface gives the AI a face. Executives warned against calling it Clippy 2.0. The comparison still surfaced in coverage.

Search in Windows 11 receives its biggest overhaul in years. The taskbar box now performs quick actions. Turn on dark mode. Adjust volume. Send a text. Results mix local files, web answers, inline previews. The update rolls to all Windows 11 users this fall. It uses a cleaner WinUI 3 design.

Security mattered. Autonomous agents create new risks. Microsoft released Microsoft Execution Containers, or MXC. The technology sandboxes agents. It runs them as separate users with strict permissions. Agents cannot escalate their own access. The feature becomes generally available across Windows 11. It works on existing machines, not just the new premium hardware. Nadella emphasized this point. TechCrunch highlighted the broad rollout.

Executives painted agents as future heavy users of the file system. The PC must serve software that operates independently. Windows gains routing logic called HydraFusion in GitHub Copilot. It decides when to use local models and when to call the cloud. Simple tasks stay on device. Complex reasoning travels to frontier models. A new MAI Code 1.1 Flash model demonstrates the approach. The 137-billion-parameter mixture-of-experts compresses to run with far less memory. It activates only 6.8 billion parameters at once. Quantization cuts size by 80 percent. Performance holds on coding benchmarks.

Meta joins the effort. Its Muse personal agent arrives as a native Windows app. The partnership expands the choices available inside the new Copilot interface.

Huang called the RTX Spark “a beast.” He reminded the audience that Nvidia was founded because of Windows. Nadella spoke of hybrid intelligence that combines local speed, cloud power, and purpose-built devices. “Just having the model doesn’t do much,” he said. The full stack matters.

Analysts question whether buyers will pay these sums. The highest-end Ultra sold out quickly in some configurations. Yet mainstream PCs still ship with 16 or 32 gigabytes of memory. The 120-gigabyte recommendation for best local model performance creates a clear split. High-end machines for agents. Everything else for traditional work.

Memory prices rose again this year. Demand from AI training pulled resources away from consumer devices. Nvidia itself raised prices on related systems by 75 percent in recent months. The Surface Laptop Ultra at $2,599 sits well above typical Windows laptops. Microsoft offers trade-in deals up to $1,000 for MacBook Pro owners. The message is direct.

Other PC makers follow fast. Asus, Dell, HP, Lenovo and MSI announced RTX Spark machines. They share the same chip, similar memory options, and the updated Windows features. Availability aligns with Microsoft’s October 16 date.

The event avoided talk of Windows 12. Changes arrive as updates to Windows 11. Yet the cumulative effect feels like a new platform. Execution containers. Local model support. Agent identity through Entra. Routing that balances cost and capability. Developers can test large models without cloud bills. Creators run demanding workloads without constant uploads.

Performance demonstrations showed real software. Video editing alongside gaming and AI inference. Gears of War ran smoothly. Call of Duty arrives later this year with full ray tracing, DLSS and Reflex support. The hardware handles mixed loads without thermal throttling thanks to improved cooling.

Microsoft positioned the announcement against Apple’s latest silicon. The numbers favor the Windows side on AI-specific tasks. Apple has its own local intelligence features. The comparison will drive marketing battles in coming months.

Longer term, the bet is structural. PCs become runtimes for autonomous software. Agents index files overnight. They prepare reports before morning meetings. They route routine code changes to local models and escalate hard problems. The operating system must protect data, manage permissions, and allocate resources intelligently.

MXC provides the foundation. It treats agents as distinct principals. Policies live outside agent control. Enterprises gain tools to audit activity. The approach addresses fears that grew as agents gained autonomy.

Early reaction on X mixed excitement with sticker shock. Some developers praised the local coding model. Others calculated total cost of ownership against continued cloud use. Memory constraints remain real. A 128-gigabyte machine delivers impressive headroom. Most users sit far below that threshold.

Still, the direction is set. Microsoft and Nvidia spent years co-engineering this stack. The event moved the conversation from concept to shipping product. Prices are high. Capabilities are concrete. Windows now ships with tools designed for agents first, humans second.

That inversion may define the next decade of personal computing. Or it may prove too expensive for broad adoption. The next 12 months of sales data will decide. For now, the hardware exists. The software is rolling out. And the agents have a home.